Pin diffusers to abot-world-modular@206121e08 (tiny VAE decode in the streaming preset)

app.py

@@ -2,7 +2,7 @@
 ABot-World on Modular Diffusers — interactive action-conditioned world rollout.
 gradio.Server + WebSocket live-backend edition.
 
-The engine is one `pipe.stream(action_source=...)` loop over the Modular Diffusers
+The engine is one `pipe.stream(actions=<callable>)` loop over the Modular Diffusers
 `ABotWorldStreamingBlocks` preset (huggingface/diffusers#14159): the pipeline polls the
 held keys once per generated block and yields every block's decoded frames.
 The serving shell (gradio.Server, WebSocket frame stream, pacing, per-session queues,
@@ -344,7 +344,7 @@ def create_gpu_rollout_loop(command_queue, seed_path, prompt_text, seed):
                 image=image,
                 height=STREAM_HEIGHT,
                 width=STREAM_WIDTH,
-                action_source=action_source,
+                actions=action_source,
                 generator=torch.Generator("cpu").manual_seed(int(seed)),
             )
             for event in events:

requirements.txt

@@ -1,5 +1,5 @@
 accelerate
-git+https://github.com/huggingface/diffusers.git@206121e08010ddc3451c3ad232ec166657a6da7f
+git+https://github.com/huggingface/diffusers.git@d2d0aeab86fd6e17c97d04266ecad5f25c168c24
 torch
 transformers
 ftfy